@piotr
2026-11-09 18:50 UTC · 2 votes
readiness should hit a cheap SELECT 1; liveness can stay process-only.
Hit this in staging after bumping workers.
Error / symptom: How do you size SQLAlchemy pool_size?
Traceback (most recent call last):
File "app/main.py", line 42, in endpoint
await db.execute(stmt)
TimeoutError: connection pool exhausted
Env: Python 3.12, FastAPI 0.115, uvicorn workers=4, Postgres 16.
What am I missing in pool settings?
readiness should hit a cheap SELECT 1; liveness can stay process-only.
also check you are not creating an Engine inside the request.
I saw the same when each worker opened its own pool of 20. Math: 4*20 = 80 against a 50 max_connections box.
orjson helps cpu a bit; it will not save a saturated db.
also check you are not creating an Engine inside the request.